Online Capacity Management for Increased Resource Efficiency of Software Systems

Resource efficiency is an increasingly important internal quality attribute of software systems. While the related attribute performance is mainly concerned with metrics quantifying timing behavior and resource usage characteristics, resource efficiency is a measure of a system’s resource usage economy. Many software systems are exposed to varying workload conditions significantly influencing its timing behavior. However, the capacity of those systems is typically managed in a static and pessimistic way, causing temporarily underutilized resources, e.g., application servers, during medium or low workload periods. SLAstic, the self-adaptive approach for online capacity management developed in this work, aims to increase the resource efficiency of distributed component-based software systems employing architectural runtime reconfiguration. A software system is equipped with reconfiguration capabilities that allow to control a system’s performance and efficiency properties at runtime in an elastic way, e.g., by migrating and (de-)replicating software components, and (de-)allocating server nodes. Architectural models play an important role in the approach since they are used to specify the system assembly, deployment, instrumentation, reconfiguration capabilities, performance properties etc. At runtime, these models are continuously updated and used for online quality-of-service evaluation, e.g., workload forecasting and performance prediction, in order to determine required adaptations and to select appropriate reconfiguration plans. A prototype implementation of our adaptation framework [1] is used to quantitatively evaluate the approach by simulation and lab experiments, based on an industrial case study system [2].